Large Language Models (LLMs) are making waves in various fields, but can they actually teach? A new study from the National Research University Higher School of Economics in Moscow puts LLMs through a rigorous, psychometrics-based benchmark specifically designed to assess their pedagogical competencies. Instead of relying on existing exam questions or generic benchmarks, researchers built a novel assessment focusing on the skills a teacher's assistant or consultant would need. They crafted original multiple-choice questions, categorized by Bloom's Taxonomy (reproduction, understanding, and application), covering 16 pedagogical content areas like classroom management, instructional design, and special needs education. The results, based on testing GPT-4 in the Russian language, reveal interesting insights. While the LLM showed some competence in understanding pedagogical concepts, it struggled when applying that knowledge to solve real-world teaching scenarios. This highlights a crucial gap in current LLM capabilities: while they can regurgitate information, they lack the deeper reasoning and problem-solving skills essential for effective teaching. The study underscores the need for more sophisticated benchmarks that move beyond factual recall and delve into the complex cognitive processes involved in education. The future of AI in education hinges on developing LLMs that can not only understand pedagogical principles but also apply them intelligently and adaptively to support diverse learning needs.